Sensor-based machine learning for workflow detection and as key to detect expert level in laparoscopic suturing and...

Area of Science:

• Surgical Education
• Biomedical Engineering
• Machine Learning

Background:

• Traditional surgical skill assessment relies on subjective expert ratings.
• Modern technology offers potential for objective, automated skill evaluation and workflow analysis.
• Laparoscopic surgery training can benefit from advanced assessment tools.

Purpose of the Study:

• To evaluate machine learning (ML) algorithms for skill level assessment in laparoscopic training.
• To assess ML algorithms for phase detection during laparoscopic tasks.
• To utilize Myo armband sensor data for automated surgical performance analysis.

Main Methods:

• Participants of beginner, intermediate, and expert levels performed suturing and knot-tying tasks.
• Myo armbands recorded motion data (acceleration, angular velocity, orientation).
• Machine learning algorithms (decision forest, neural networks, boosted decision tree) were compared for skill and phase detection, with and without Dynamic Time Warping (DTW).

Main Results:

• A neural network regression model showed the lowest error in predicting Objective Structured Assessment of Surgical Skills (OSATS) scores.
• An ensemble of neural networks achieved the highest accuracy in predicting skill levels (82.2% for beginners, 79.5% for experts).
• Phase detection accuracy improved from 16% to 43% with Dynamic Time Warping (DTW) and boosted decision trees.

Conclusions:

• Machine learning algorithms can interpret complex surgical motion data for skill assessment, outperforming standard statistical analysis.
• Dynamic Time Warping (DTW) enhances automated surgical workflow detection by processing motion data.
• Further research is needed to standardize data interpretation and improve sensor accuracy for wider adoption.